Research On Recommendation Technology Based On Non-Compensatory Decision Mechanism

It is well known that recommendation systems using machine learning methods have become one of the most active fields of data mining.Recommendation Systems have been successfully applied in E-commerce and other fields.An effective recommendation system can greatly increase a company's sales.Consumer psychology reveals two consumer decision-making processes:compensatory rules and non-compensatory rules.The existing recommendation models based on latent factor models assume that consumers follow the compensatory rules.In compensatory rules,consumers evaluate an item from multiple aspects,and calculate the weighted or sum scores of various aspects,which are used to obtain the score of an item.However,many literatures on consumer behavior indicate that consumers use non-compensatory rules more than compensation rules.This thesis mainly studies the recommendation technology based on non-compensatory decision mechanism and the derived model in the recommendation model.Based on the non-compensatory rules,this thesis proposes a conceptual model of how users use non-compensatory rules in the recommendation system.Based on the proposed non-compensatory model framework,this thesis applies it to various existing recommendation systems,including rating prediction model and ranking model.Experiments show that the use of non-compensatory rules can effectively improve the different models'recommended performance on various real datasets.Furthermore,based on the non-compensatory model proposed above,the strength of implicit feedback is introduced in this thesis.In order to exploit implicit feedback to recover user preference,we propose to incorporate the strength of positive implicit feedback to optimize recommendation and simplify the recommendation algorithm inspired by negative feedback's uncertainty.We also propose the prediction strategy based on the ordinal utility.Experiments on standard benchmarks demonstrate that the proposed methods enhance recommendation performance.